Agentic KPI Tracking Skill
Guide measurement and tracking of agentic coding KPIs to assess ZTE readiness.
When to Use
- Measuring agentic workflow effectiveness
- Tracking progress toward ZTE
- Analyzing success patterns
- Identifying improvement areas
Core KPIs
Summary Metrics
| Metric | Calculation | Target | | --- | --- | --- | | Current Streak | Consecutive successes (Attempts <= 2) | Higher is better | | Longest Streak | Best consecutive success run | Track improvement | | Average Presence | Mean attempts across all runs | Target: 1 | | Total Plan Size | Sum of all plan sizes | Track scaling | | Total Diff Size | Sum of all changes (added + removed) | Track throughput |
Per-Run Metrics
| Metric | Source | Meaning | | --- | --- | --- | | Attempts | Count of plan/patch runs | 1 = perfect, higher = retries | | Plan Size | Lines in plan file | Task complexity | | Diff Size | Lines added + removed | Change magnitude | | Files Changed | Number of files modified | Change scope |
Calculation Methods
Attempts Count
Only count workflow restarts:
attempts_incrementing = ["adw_plan_iso", "adw_patch_iso"]
attempts = count(workflow in all_adws if workflow in attempts_incrementing)
Build/test/review don't increment - only full replans.
Streak Calculation
current_streak = 0
for run in reversed(runs):
if run.attempts <= 2:
current_streak += 1
else:
break
Diff Statistics
git diff origin/main --shortstat
# Output: X files changed, Y insertions(+), Z deletions(-)
KPI File Format
Store in app_docs/agentic_kpis.md or equivalent:
# Agentic KPIs
## Summary
| Metric | Value |
| --- | --- |
| Current Streak | 5 |
| Longest Streak | 12 |
| Average Presence | 1.3 |
| Total Plan Size | 450 lines |
| Total Diff Size | 2,340 lines |
## Detail
| Date | ADW ID | Issue | Class | Attempts | Plan Size | Diff +/- | Files |
| --- | --- | --- | --- | --- | --- | --- | --- |
| 2024-01-15 | abc123 | #45 | /bug | 1 | 35 | +45/-12 | 3 |
| 2024-01-14 | def456 | #44 | /feature | 2 | 85 | +120/-30 | 8 |
Tracking Workflow
Step 1: Gather Current Run Data
From state or git:
- ADW ID
- Issue number
- Issue classification
- Plan file path
- All workflows run (for attempts)
Step 2: Calculate Metrics
attempts = count_attempts(all_adws)
plan_size = wc_lines(plan_file)
diff_stats = parse_git_diff()
Step 3: Update Detail Table
Add new row with current run data.
Step 4: Recalculate Summary
Update all summary metrics based on full detail table.
Step 5: Analyze Trends
- Is streak increasing?
- Is average presence decreasing?
- Are plan sizes growing (handling bigger tasks)?
ZTE Readiness Indicators
Based on KPIs, assess ZTE readiness:
| Indicator | Threshold | Status | | --- | --- | --- | | Current Streak | >= 5 | Ready to try ZTE | | Average Presence | <= 1.5 | Good efficiency | | Recent Failures | 0 in last 10 | High confidence | | Plan Size Trend | Increasing | Scaling up |
Key Memory References
- @agentic-kpis.md - KPI definitions from Lesson 002
- @zte-progression.md - How KPIs relate to ZTE levels
- @zte-confidence-building.md - Using KPIs for confidence
Output Format
Provide KPI update:
## KPI Update
**Run:** {adw_id}
**Issue:** #{issue_number} ({issue_class})
### This Run
- Attempts: 1
- Plan Size: 45 lines
- Diff: +67/-23 (4 files)
### Updated Summary
- Current Streak: 6 (was 5)
- Longest Streak: 12 (unchanged)
- Average Presence: 1.28 (improved)
### Analysis
[Trend observations and recommendations]
Anti-Patterns
- Gaming metrics (easy tasks only)
- Ignoring failures (not counting retries)
- Not tracking consistently
- Celebrating streaks over actual delivery
Version History
- v1.0.0 (2025-12-26): Initial release
Last Updated
Date: 2025-12-26 Model: claude-opus-4-5-20251101